Neural network based Fault Tolerant System for Cascaded Multilevel Inverters
1. Understanding the Fault Tolerance Concept
Overview of Cascaded Multilevel Inverters
Cascaded Multilevel Inverter Configuration
For our 15-level Cascaded Multilevel Inverter (CMLI), we require seven H-bridge inverters (H1, H2, H3, H4, H5, H6, and H7).
Each H-bridge is responsible for generating a specific level of output voltage.
Types of Faults
Inverter Failure: Complete failure of an H-bridge inverter.
Switch Failure: Individual switches within the inverter may fail.
Battery Failure (DC Source Failure): Issues with the DC power source affecting inverter operation.
These faults need to be detected and corrected to maintain stable output voltage.
Fault Detection and Correction
Neural Network for Fault Detection
A neural network will be trained to detect faults based on the voltage measurements across each H-bridge and the load.
The network will identify which specific H-bridge is failing and trigger appropriate corrective actions.
Corrective Actions
When a fault is detected, a backup H-bridge inverter will be activated to compensate and maintain the required output voltage.
2. Implementing the Fault Tolerance System
Simulink Model Setup
Building the Model
Create a Simulink model with seven H-bridge inverters.
Incorporate ideal switches and constants to simulate normal and fault conditions.
Fault Creation
Implement fault conditions by changing constants to simulate inverter failures.
Include an additional H-bridge inverter (Aary H-bridge) that will be activated during fault conditions to maintain output voltage.
Data Collection for Neural Network Training
Gathering Data
Measure voltages across each H-bridge and the load under normal and fault conditions.
Collect data for various scenarios, including normal operation and faults in different H-bridges.
Preparing Data for Training
Create a dataset with both normal and fault conditions.
Generate a significant amount of data (e.g., 10 sets of samples) for effective neural network training.
Labeling Data
Assign labels to the data indicating the specific fault or normal condition.
For example, if a fault occurs in H-bridge 3, label the data accordingly.
3. Training the Neural Network
Neural Network Training Process
Setting Up the Neural Network
Use MATLAB's Neural Network Toolbox to train the neural network.
Input data includes voltage measurements, and output data includes fault classification.
Training and Validation
Train the neural network using the collected data.
Validate the network to ensure accurate fault detection and classification.
Ensure that the network output matches the expected results for various fault conditions.
4. Simulation and Results
Running the Simulation
Simulating Normal Operation
Run the model with all toggles set to simulate normal operation.
Verify that the system outputs the correct 15-level voltage.
Simulating Fault Conditions
Introduce faults in various H-bridges (e.g., H-bridge 1, 2, 3, etc.).
Observe the network’s ability to detect faults and activate the backup H-bridge inverter.
Results and Analysis
Fault Detection
The neural network accurately detects faults and activates the backup inverter as needed.
The output voltage remains stable at the required level despite faults.
System Performance
The fault tolerance system maintains the 15-level output voltage even when faults occur.
The neural network effectively identifies faults and ensures continuous power supply to the load.
Comentarios